Why MDR Technical Files Keep Breaking Down

Hey everybody, welcome. My name is Sean Smith. This is MedTech Leading Boys Exchange Friday and Focus Call. I am, reporting live from Boston, Massachusetts today. And we have two guests, from CertHub, Leon Gobinger and Nicholas Gehring, and they are going to be talking to us about. Mdr technical files going off the rails. And all the different ways that can… that can happen. While we're waiting for people to join, I would like to mention, next week we're gonna have Suresh Suresh from, Tusood is going to be talking about in-silico testing,

he's a doctor, he's a lawyer, he's, got a lot of knowledge, he's an auditor, he's a clinician, and I think… and he's also got a lot of interesting opinions about… in silico. So I'm really, really looking forward to that session next week. And then… Mid-month on June 17th, I'll put the links in the, In the chat, we're gonna be doing a webinar with, Claramed and Canyon Labs about, packaging pitfalls, which, is a super important topic, and hopefully you'll be able to join us for that. As I said, I'm gonna put those in the thing, in the chat.

I'm gonna try not to talk a whole lot today, because I'm in a semi-public place, and but, I will, I will chime in as needed. Let's see, where did I meet Leon and Nicholas? I met them… most… in Lisbon, correct, guys? Correct.

Correct.

And, they were, exhibiting, and I was, going around and meeting all the people that were there. Now… As little children growing up in Germany, did you dream of one day owning CertHub?

I mean, doesn't every child wake up in the morning and want to do a medical device compliance? No, no. Just joking, but I think, you know, when we grew up, I think both Leon and I were really… fascinated with, like, technology, and also deploying technology in highly regulated industries. Leon can also tell a little bit more, because he… he's basically been his whole life in highly regulated industries. For me, it was a lot with AI. Yeah, and it took us a while to find each other, but eventually we did, and then we decided there has to be something better out there, and… that's kind of the mission why we founded Search Hub, and yeah, it's been a hell of a ride and a lot of fun in the last years, so also happy to be here today.

Yeah, I can just also add on to it, yeah, I think for me it was similar. I jumped, actually, in a very young age already in this highly regular industry. I actually started in aerospace, did, like, system engineering for air graphs, flight testing, QMR in the aerospace field, and then actually let me think what year it was… 2019? 2018? To be honest, also because of my girlfriend is a medical doctor, and then I got this really big excitement of medical devices, and I thought. why can I not use this, you know, system engineering skills and this quality management record affairs skills, maybe in a different industry? And then I did my medical informatics master's, and…

Then I went into med tech companies, did certification support there were, like, QMB, PRC there, and… yeah, and then came really into this industry, and I honestly have… my heart is in both worlds, kind of, in med tech and aerospace. This is, like, the industries I love, and… It's hard sometimes, but it's also really, really exciting products and solutions there.

So, before we jump into the meat of the session and let you run your thing, I want to talk a little bit about the big picture, where we are as an industry in terms of digitalization, in terms of digital transformation, etc, and the ability, To access and share data and information fluidly. Where, you know, on a scale of 1 to 10, where would you say the med tech industry is?

Nico, you first. the honest engineering,

I think…

Assessment.

With these things, right, there is no uniformal answer. I think it is very, very different depending on the type of companies, and also how companies see regulatory as a strategic field. You know, there is always, like, these two… these two divides between people who see it as just a burden I have to do, and they don't invest in it, and then you have companies who really understand market access, being compliant, like, the cost of quality also. And they invest very heavily in these things, because, I mean, it's the age of AI, we all want to reap the benefits, but to do that before, you also have a, like, significant level of digitalization. So, Sean, I would say I see…

Between 3 to 7. this is, like, the range where I would put most of the companies, and especially when people start new developments, that is a perfect time where they adopt, like, new strategies, and then they are more on the upper end. But I think it's also interesting what Leon is observing, right? There is also a big difference between the industries. Between highly regulated industries.

Yeah, maybe… yeah, because my point, I can… I can better compare it with the back end of work with aerospace. I think PLM and product lifecycle management, the engineering part, is far more digital. It was really tracing through tools there, structuring, since many years, but… On the quality managed record affairs part, to be honest, there was… now you need to watch out where you are. Are we in, we are in the space section? Are we in the military section? Are we in the civil aircraft section? Especially in space and military, there's far less standardization, I would even call it, and, even less standardized process to get there. I think there, I think MedTech is in a good path.

regarding digitalization as well, I think I'm with Nico there. What we see in daily working with manufacturers, we see everything from I would even say from 2 to 8. Like, the age… the times where you really work on paper paper, physical paper, it's getting less and less. I think we had a few customers at the start which really were on this level. But this is the range I see, personally. It's… so, for me, the most interesting part is that the range is so huge.

So, why don't you guys move ahead with your presentation, and walk us through what we're gonna see today, and, yeah, I'm excited about it.

Cass as well. Then Leo, now it's your time.

Perfect. So, what we wanted to talk about is, why technical files keep breaking down. I extended it because of some, also. and your US participants today, also global submission in general, so not only about technical files, and we wanted to share some reword challenges we see, and just solution proposals regarding this. And solution proposals, not theory, we also show, so on the software, like, what… on ours of how we approach things, is this the ground truth? Are there different approaches for that for 100%? But also, one interesting thing is also, we talk so much these days about AI solving XYZ, AI solving this, and we want to also put, like, our opinion on that is…

why we need to do one step before to actually leverage some of them, and this is something we really want to emphasize, not because we like this, we would love to just use our cloud for everything, but we also make… tell you… tell a little bit where we see where we are getting to the limits with generic AI systems compared to domain-specific ones. This, you already know, just a little face to it, I always love also to… as a poor person, so my… maybe the only one thing I want to mention is, like, the superpower for me, it's really important. Systems needs to be easy, usable, it needs to be fun to work with, and regular to compliant in the same way. And sometimes it feels, like, contradicting to each other, and this is maybe my superpower, I really want to merge these two things.

Together.

Perfect. That's me. I think I also had a chance to introduce ourselves. The one thing I want you to take away, I spent a lot of, lot of time here at Technical University of Munich doing AI research, really figuring out how you can use these systems to actually do legal reasoning, do compliance checks, and… You also see some nice pictures of us, we are not only here working, there is also a person behind it, so… We spend a lot of time in the mountains when we have some little time left, but, yeah, that's about us, but I think let's get into what we want to talk with you about today.

Exactly. So what's our agenda for today? We do a short introduction, just like, who's this company around us to, kind of, then a motivation regarding it, why actually do we solve the problems we are solving, and some current challenges. And the rest of it, like, that's… I would say the theoretical part behind it, 10 minutes. The rest is, we go with one problem we see, and go with one solution, and then we show the solution in the software, so 2 to 5. Are really also looking into solution, approaches to it, and the last one is the future outlook. We really try to be also fast and efficient here to all answer your questions, and please write them in the chat, so always the person not talking, try to also check and tries to answer your questions.

Exactly. So, company-wise, we are, based in South Germany, Munich, we are 20 people at the moment, maybe what makes us unique, we have a cool mix of, software engineers, technology-driven people with regulatory experts in their team, and this combination is actually what is very fruitful in our team. We are, supported by a lot of experts in the fields, we have a lot of partnerships also with regulatory consultancies partnering on it, and, are… Yeah. I think that's the most important thing to know. Where do we fit on? we fit between the product developing and the access to the market. This is where we have our sweet spot, where we love to be. It's always, like, regulatory relevant information coming to our system, and we bring it into a format for submission, for,

the notify body that it's fitting, to the formats they require. And there, we also have normally two ways to do it. Sometimes we have companies who, getting their first certification, and sometimes we have companies who, since 70 years on the market, who really are, like, 50 years on the market, have 70, 100 product variants. selling in 10 USD cations, and this is the focus for today. We call this the one-click documentation. We don't talk too much today about this fast track, there are other webinars regarding this. Exactly. Solution-wise, as I mentioned, we do the one-click documentation topic, you will see it in the software later, we don't need to talk too much about it, and we have assessments, so, like, pre-assessment before you make the actual submissions, either to notify body in the EU, or also for FDA 510K, de novo, or if you're doing a visa submission.

Giving certain pre-checks. And this can be on a documentation level, or on a QM audit level. Exactly. Maybe what's very important is that we don't… normally don't have, like, an empty system you get, for example. We normally always… this is, like, one thing I really love. what we're doing is, we tailor our device… our broader, or the software, to the type of medical device of the customer. So if you're an IVD device, you get a different configuration than if you have an embedded system, if you have sterilisator, which I will talk later about. If your software as a medical device, and we can configure this kind of automatically in the system.

And we show you why this is actually very helpful. Enough about us, let's go to topic of today.

Cool. So, why do we do all these things, right? In the end, I think people also attending the call, we deeply care about the products we built. We want to bring it to the market, we want to bring it to the patients, and what we see really as today's dilemma is really this. ping pong, right, between you as a manufacturer, taking a lot of time, assembling the documentations, sending it over. In Europe, we call them the notified buddies. In the US, you have the FDA. But this whole process takes too long, in our opinion, and a lot of opportunity is lost. So what Leon has not mentioned, but as a quick reminder, we also partner here with a lot of European notified bodies, and FDA has developed their ELSA tool, right, to speed up the checks. We do the same thing here for the European ones.

And now that the notified bodies are also getting quicker, we also need to make the manufacturers quicker. So in the end, our big vision is to make sure that new products enter the market faster, and existing products stay in the market easier. So that is the main motivation why we are doing what we're doing. Exactly, and now we are coming to our learnings, and also to the things which we see in daily life, and in the projects with a lot of people. And here, I'm also curious, and people listening, if you want to interact, please feel free to also post a bit in the chat if you agree, disagree, what of these problems you're facing.

But from all the projects we have, really what causes the most headaches is that people really struggle with keeping all their documents they in the end, submit and create. Error-free, redundancy-free, for example. your intended user, or your patient group. You repeat these types of information in so many documents, and then you update it, and you usually forget one, you usually have inconsistencies between what is written in your clinical file versus what the latest version in, Another statement is, and that really, really results in a lot of manual work. Right? The underlying reason, in some way, is also that people really get lost in the traceability.

I mean, we know a lot of things have to be connected, right? You need to trace from your initial user requirements to your tests. To maybe the kapas, you know, who, in the end, caused this new user requirement, and… Because of the first thing. the errors and redundancies, right, the traceability kind of just magnifies this problem, because you change stuff, and then you really, really get out of control. And then, the last stuff, we call it the portfolio explosion. Or sometimes. it's some… we jokingly also call it a bit the burden of success, right? The more products you're then having, the more markets you're selling to. All the aforementioned problems

they just explode, because now you update one thing, and you don't only have to update it for your EU submission, now also in the FDA documents it's out of date, and in your own visa ones, so… all the problems grow exponentially, and this is where we see the current state of the industry, that QMRR people really, really struggle with these type of things, and especially when it comes to changes, so change management as the underlying thing. Exactly. 10? This is also, I think, Sean, it fits very well to what you have asked us before. From 1 to 10, how do we see it? We have our level 0 to level 4 internal framework, I would call it.

And why do we show it here? I think… People know that digitalization is a solution to these things, and AI will also help you solve these problems. But these are so big of a buzzword that We sometimes really have to explain people what level of digitalization they can actually have and reach, and I want to go through each of the levels, and as people feel, like, interacting, it would also be very curious to hear your self-assessments on these things, because Level 0 is really the paper. This is when you are working with physical paper, you're signing with wet ink, you're putting it in real physical folders.

This we don't see so much. But then, level one, this is… people have digital PDFs, people put the data in SharePoints, people are using DocuSign to digitally sign it, but still, a lot of stuff is managed in a, like, yeah, digital administration, right? The physical document is now just a digital one. Level 2, this is also where most bigger companies are at the moment. This is where we see these classical document management systems. Right? Then you take your digital document, but you put it in something like a wind chill, and then you have the option to version control the document properly. You have the option to trace between these documents and approve them.

And now comes the important shift, and this is where we advocate for. It's the Level 3. We call it the data-driven documentation, and there is a mindset shift. Because we say. the documents, in the end, they are made for humans to look at, right? And all the problems with the redundancies and repeating ourselves comes Because we put the same information there multiple times. So why don't we shift our focus on the regulatory data, right? Think of a big database where you have each important information exactly one single time, the single source of truth. And you only work and approve on this data, you make sure that this is traceable, and the documents, the documents for

Any jurisdiction you want to sell today, fall out of it, if you have the right template. So, this is the level where we want to see more companies at, and this is also what we see by working with a lot of stakeholders in the industry, where this is going, and also what Brussel expects from us in the future. And the level 4, and why we also call it the AI use cases, everybody wants to use it. And we see you have to reach level 3 in order to really, really reap the benefits of the modern AI technology. And we are here today to advocate for you for Level 3 and 4, and… Exactly. This is, how we currently see the market and the different challenges.

Perfect, thank you, Nicholas. And that's, like, level 3, level 4, it's all about the rest of the presentation. This is the only thing you need to remember to understand the… For the rest. Good. So, on the reword examples, let's go into practice. So, the first problem was error and redundancies. So, as you said, we have different types of the documents, we have an intended purpose who is cited 50, 60, 70 times in different wording, someone forget to change an information there. I think everybody can relate to this. So the solution idea is now, yeah, cool, let's lose a structured data model, we have single source of truth, we have end-to-end traceability, we have a road-based approval rights, now we have all these advantages, but what is the big…

downside? now I need to have a huge digital transformation project, I need to… I'm currently here, I have my SharePoint, all my documents there, how do I get this into structured format. Because this is, my personal experience, one of the main points why people say, yeah, it's cool, but I don't have the time, I need… I have… an audit in 2 months, I don't have time to make this digital transformation. I don't have the capacity in the team to do it, and I want to show you today a little bit, like, what is possible to use new technologies to make… speed up to go to this level 3, actually. And this is what we call the instant we transform from legacy document into structured data. So really saying, like, I just use my current documentation and extract it into the right format.

And now, if I have it in a structured format, like, I have a patient population with a certain age group. then I can submit it to all the documentations, or to all the submissions I require at the end. And the only thing I need to know, is this information for a European submission, is this information for a US submission, or maybe are they the same, because we have a lot of overlap, regarding them. And this is what I want to now show you also in the live demo. One second… Let me just jump into it… Good. So, regarding the… what you see here. So now we go into the product, so just that you have a little bit of a navigation, you see here the overview, you see here the requirements lying in, you have the product section, you have the QMS overview quality management system. We skipped for the sake of time today a little bit,

But, let's go really into the products. So, regarding the products itself, we have an example serverly SATA system here already, and now… we set up a new system, which we call Create a Product. We now say, proceed with guided setup. And now… for the sake of this demo, I created a very short, like, website, just explaining a sterilisata as an example device. Obviously, in reality, we would submit… import here an actual full tech doc of a customer, but obviously… obviously here, this is a… Is it generated? This is not a real medical device here. It's just for the sake of showing it. Then, I can now click it in, and now it can extract from the website, or what I later show you from the document, the right information. So, what is the intended medical purpose of the device.

what is the target patient population, the clinical environment, and this is the single source of truth Nikos mentioned. And overall, we have depending on the type of device, around 1,000, 1,500 of the single information. And they are now subtracted in here. Good. Now I go to the questioner. Now I have a guided questionnaire. If I don't know what type of device I have, what I need to structure, classification, then I can singly go through the guided questionnaire. I just make the fast lane shortly. say we are an MDR, we are sterile, we have an active device, and why do we do this? Because now we only upload data items which are relevant for my type of device. This is what I mentioned earlier, like, I don't want to have a performance evaluation in there, because I'm not an IVD in this case.

So, in the classical world, this would be if tailoring automatically templates to the device. So, we have a Class B device, for example, and let's say we make HN studies. Class 2B? And now I click on… few… so these are my product properties, and now I set up the device. And set up the device means We at this moment skipped the SOPs and template for it. Let's see, now we set a 20A, let's make the example. And what's now happening in the background… we now set up the… I call it always the LEGO… like, I always think of Lego bricks, and I built the house of my LEGO bricks now, putting the single building blocks in there. And to make you understand a little bit what this LEGO bricks is about, in the time it's loading, I go into an already filled one.

The stericata 23B? And there you see the device description, device specification, clinical evaluation part. And you already see an estimate how much of the information are already filled. So, we have information supplied by the manufacturer, we have post-market surveillance topic, product developing, risk management here, there's also requirements engineering in there, usability engineering. Let's make a simple example. Let's go for device description. And here you see intended use, general description, application, and you only have this one time in here, the patient population with the single building blocks.

And normally, someone only needs to fill out this information. And now, let's see if it's set up already. It's a few seconds more. Because what I want to show you…

interrupt you here, but I think I really want to emphasize this. We have… it's an… immensely huge data model we have built and also researched, together with… there is, like, a effort here in Europe, the Medical Device Knowledge Unit, where we are also, like, pushing a lot the standardization. There's a dean spec of this data model, and we are now loading these 1,500 tailored data points given these product properties Leon had shown before, as you can see it. And we also trace them, and if you want to, we set up also the correct SOPs and templates for you.

Exactly, and now, I could do it here, but I show it you in the full… oops… Skip to next, just ignore this. Here, I can now… upload an existing document where this information already lies. So, it's called FillDeviceDatabase, because obviously I showed you a full one, this one, obviously, is empty now, because we only set it up now. Now, I go to FillDevice Database, and now I have here… a device description or specification of my not-real-existing medical device, kind of, or example. But obvious, this could be any format of any structure of a manufacturer. So this is normally what you know from the structure-wise.

So, following Annex 2 also of the MDR. Slow this in. And now, in the next step, I can now select which information I want to fill. Let's make the device specification, for example, and the device… Where do I have it? Device description here. Exactly. Now this is processing and extracting this information out. In the time it's extracting it. I can also show you something else, which comes later, another problem, I just want to show you now, so we don't need to jump back again. Now? I can make…

the change management Nico earlier mentioned. I can, for example, change the name of my device, call it 23A, and now run an impact analysis on it. What are the impacts Of this, title change, to which of the submissions, to which of the documents, does this have an impact? And here I see all documents which are linked to this product. Yeah, there's a lot, yeah. And we also have… all the submissions line, then, and… and I can also get an AI assessment, which is running here, to say, like, what is the severity of the impact. So sometimes I do the example, if I change the intended purpose, I mean, that's very simple, because it has huge impacts, of course, but this you can do everywhere in the system.

So the same I could also have done in… Going into… into the… into a specific, for example, the features of my system. I could call it here, High precise temperature control? as our high precision temperature control, I update it. And now I can also say, like, impact analysis again. What is the implication of this change? Which documents, which submissions? And also analysis of the actually con… like, does this… it analyzes the content and says, what are the implications of that? Exactly. So here we solve topics of traceability combined with a clever or smart change management.

Good, let's see if it's already filled out. Perfect. So now I can go through the single building blocks and check what is already filled out from the single parts. So, for example, I go into, because it was device… description, for example, it automatically filled out the intended use statement, it already filled out, so I save it, I say, that's cool, that's good. I fill out the patient population, there was some information not in the document, which obviously is no problem, I can change now, simply, or also later, that's no problem. It's just… it's not always the… it's not the insurance of completeness, it's time-saving is the main advantage of it. So you fill out 80% automatically, and the rest, you need to check, and that's always what I always say. AI is super helpful to speed up things, you need to understand what you're doing.

This is, my humble opinion regarding it. And here I can also accept it.

maybe quickly interrupt you, Leon, also giving the questions in the chat. I think what is very important to emphasize here for us, because the question was, how do we measure accuracy, how do we measure, or like, how do we keep the quality up to date? So, whenever there is anything with AI usage in the system. there is always this human-in-the-loop approach, right? There will never be a scenario in the system Where, the AI is just doing things on a.

For example.

Without a human correcting that, and… how do we do it and develop it in-house, right? We have a lot of internal benchmarks, doing these extractions, figuring out how good we are, and we are constantly improving our own models and our own systems to turn exactly what you're seeing here, this instant, transformation, into, like, the easiest and fastest project, you… you can use to, to… Reach this level 4 or level 3 we have shown before.

Good. No? We showed this with device description specification, but obviously we can make this as a bulk upload and transfer out total technical documentation in there. So, literally, if the data is in really good quality, already structured in the template, we can actually extract this super easily, and this is also why sometimes people ask us how long does this take? It actually depends how good your how redundant your information are from the start, and how good… well-structured your tech implementation is today, then this can actually work really fast. So we had, like, also a customer who submitted in a few days and transferred this completely, but it also, for some, took longer, because they… it's also a good time to also think, why do we have

Structured it in a way we did. Now, the last part of the demo for this section is now we have all this information in a structured data model. Now you say, yeah, great, super. But what do I do with it? What is the purpose for doing it? Because now. I can what we call the one-click documentation, and it really is one-click documentation, it's not a buzzword. I can say now, where do I want to submit this information? Do I want to go FDA? Do I want to go Health Canada? To a visa? I go to a single submission. I create. Let's do a new submission. It's totally set up. Example? And now, I just tell in this demo account, it's like, we have these two examples in, let's look… let's use MDR template. I go to link products, now I choose the one who's completely filled, which is a 23B.

Juristication, that's fine. So again, let's do MDR, and now I click on Create Submission. What's happening now in the background is that the information are automatic, like… the, we have, like, the obvious document templates for the… for it. They're automatically filled with the right information, so when I go into it, into a single one, it's already filled out correctly. So you see, one submission is completely ready. over here, with… sorry, not too fast. Device description, information supplied by the manufacturer, design manufacturing information. DHFs and structures work the same principle, so it's not only NX2, for example. Design and manufacturing information, general safety performance requirements, all those information are automatically filled. And you get also here

usability engineering, and you can also include additional ones. That's very important, it's not a static system, you can build your own templates as you like, what fits to your case, so it's only I always prefer to start with something existing and build on top of it than starting from zero, but obviously you never need to use our templates, and some manufacturers are also using their own templates, but use our structured model. And null? Again, in the submission analysis, you see here, I now see how many of the documents are approved, how many are unapproved, how many are deleted, and on a… this is on a document level.

So I can also… Show unapproved documents? And I also can show the missing product references, so these are the single source… the single legal bricks I need to change, so I see here, I didn't put the information into the warning section, for example, I did into the restrictive use, and I can just make a right-click and be directly there again. Good! Let's jump back into the slide shortly. So this is what I just explained. Nico, should we jump over this, just for the sake of time a little bit?

Yeah, if you want to. I think the next slide's the real message, really, for us, is… you remember the star on the pyramid? Everybody wanted to do AI, and This is here really to emphasize, if you work on the structured data, your AI results and whatever you want to achieve with it will become so much better. Right? And this is just an illustrative example, like, if you give your AI documents, it's very hard to assess it. And in the background. What we do with this data model is really build an interconnected system of your individual information, and that works so much better. And, if you go to the next slide, and I think, Sean, we also share this deck if people want to learn more about this.

this is not only our own experience, this is also very, very well proven by a lot of research, and if people are curious about the real benefits of structuring your data for AI use, there are also some sources where you can dig in. Exactly. I think, yeah. This… let me quickly just say something to this, because Our main problem number two, is this change management, right? As we have established before. And we call it bidirectional change management, because what we observe in practices Either changes happen from within your company. you update your product, or you update certain processes, and then you use this AI impact analysis to do this forward trace, kind of like saying.

these documents are outdated, I need to redo this type of risk analysis, these submissions are outdated, so from the data to your documents. And the other part, we will also show you later how we do these regulatory assessments, is when the change happens outside of the organization. Right? Guidelines update, new harmonized standard, and then you're suddenly in the position having to figure out the other way around. What do I now need to do differently? And for that, you want to run the gap analysis. This is the next way to solve this problem, so you are not, yeah. Trying to find out what has now, what has now affected what needs to change, given the update in the regulation.

So for this, again, if you have the data, if you have it traced through from submission to documents to requirements, both of these changes are becoming, basically, no-brainers. So that is a really, really cool benefit, and the thing we see that the customers like the most in the daily work.

No, the… the problem 3 finding and updating information. I mean, we talked before with the portfolio explosion already, but now it comes in, like, okay, great, now we have this level 3, what can we do now with it? And there, it's actually something we really say is the topic of when is it good to use a general-purpose AI for it, and when it's reasonable to use a domain-specific system for it. And, like, we are practical people. We think there's a lot of which can be used, where you can just use your co-pilot, your cloud, your open AI for… it's… we skip IT security completely for today, because otherwise we would absolutely say, no, please don't do it, either… only if your organization is ready for it.

Let's skip this for now, we could talk also one hour just about this, of course. Here it is, drafting easy, brainstorming myself. We use it as a company ourselves, also, in a safe environment, those things, as well, where it works really nice. It comes very poor performance if it's really domain-specific things. If the context engineering is not deep enough. We see these systems very often strongly failing. And this is in a conformity assessment way, this is an already-ready output. If you have a lot of files, it not only costs you a lot of tokens, but how reproducible are the results, because you don't give enough context to it, and how does the…

how does the system know if there's a new version, a new approval step to it, for example? So, I always see it, I dream of, like, a system where you just… really easy usability, you can easily chat with the system, but you have this regulatory compliance, you have an audit trail, you have traceability, you have all what is required, what you need to have, but it doesn't need… it needs to be hard to use, and I think it's not contradicting, and this is why I… now jump into this, how can we now chat with our quality management system, or our tech talk system? And this is, again, live demo part, so let me shortly go through it.

So let me go into the sort of brain. So this is the… this is the brain. Well… I already prepared one question, I mean, I can run it now completely again. I just asked him, can you make a comparison of different versions of a standard, of 35.5 as an example? Obvious, it makes a really nice gap analysis regarding this. I would say, if you're really nice context and prompt. Claude system, or any generic system, you could get similar results. So this would be just… the overall quality in itself is, of course, better, but you could get a reasonable result out of it. But now, let me show you what I ask as a second question.

Can you please check all my SOPs if they fulfill a newer standard of the 35.5. And then, I have a complete… it runs through my, actually, quality management handbook, a quality management system I have in this system, and makes the analysis behind it. And I can also click into it, so the nice thing is I can click in and directly go to my single processes. And the same thing, so… Just going through it. The same thing… I can also do… in Claude itself. Oh, sorry, I need to share my other screen. Here, this is my normal Claude system, and now it can call, it's called MCP.

It can call our setup data system, our platform, What's that? Oh, sorry.

We can't see your plot.

Sorry, now you can see it. Here, I called the remote MCP. This is a normal cloud instance, and I ask him, how many SCPs do you have in the system? I ask him, again, this comparison of it. And get this information out. I also asked him, can you please check all SOPs and if they fulfill it? And now I can also run through it. But there's a big difference between both of them. Do anybody can think, like, the one was in the startup system, the one was in external cloud? Can anybody of you think about what could be a… A difference between them?

I have an idea. Maybe where does the data go? Leon, do people just want to have the slides, so we will tell them the answer?

Good. So the answer is the difference is, if I access it via an MCP, via my GLOT, I need to ensure that this cloud system is, not using my actual, company data and information. So, within the SERTA brain, we can ensure this we can, as I said, talk an hour about it, how we built this, that we can ensure this data still stays with the manufacturer. in the MCP, the manufacturer needs to ensure themselves that it's possible to use it, so if you have a really safe co-pilot running in your manufacture, you could use it completely with the co-pilot chat, but there you need to ensure it on the sort of brain, we can ensure it.

That's the big difference, because the one lies in secure servers, the other one, we need… this is out of our control. Good. Then we have a very last share again. I have a very last one example, a very, one second… no. Moniko has one.

I have one. Yeah. We told you four problems, 3 we have covered so far, and three are a lot about the one we've seen so far, it's all about, okay, how do I assemble the documents? How do I make sense of the exploding complexity of all the information? And then, when you have this figured out, The last question, as always. Okay, but… am I compliant? Right? So far, we have made sure that data is always up-to-date in documents, and you assemble documents well in submissions, and we can tell you, is everything approved? Is every data item there? But now it's really about Is the information you provided actually enough?

to uphold FDA assessment, uphold a notified body review, or an internal audit. And for debt, The solution is also, again. an AI system, and because you have the data, because we have the structured data, that actually is really cool. And what we do is… this is also what we work together with, the notified Buddies on, to… on the one hand side, speed up their assessments, but also on the other hand side, speed up your side of the whole thing, to give you more, more flexibility. And now it's my turn to show you how that works. Oh, no.

of share.

It's the classical, I have to restart my Zoom, because.

Nico, I can share, I have it open.

Okay, then maybe you share, otherwise I'm gone. Perfect. Lauten, let's see, let's see, maybe it works? Freigiben? Oh, can you see my screen now? Yeah, is it working? Perfect. Then I don't have to start Zoom, so Apple did a good update. So, this is the last thing. The conformity checker. It's a thing we can do, and it runs either on the structured data. Or, it runs on your existing documentation. And the most important thing to understand here is, we can run… arbitrary assessments against that data and give you this big gap analysis. So, for example, here, we have all the different requirements and how to assess them for ISO 3485, right? So we can do a QM audit for that.

Same, we can also do something similar for technical documentation, so we go and check the MDR annexes. So, these are the tailor-built agent systems, right? This is what Leon said before, a domain-driven AI system, not a co-pilot. You couldn't do that in a co-pilot, because you can't throw in… 500 gigabytes of data, so there is a lot of things we have to do on our side. And what we then usually do is, Maintain technical documentations, or in this case, someone sent us over, their real files, right? So we read that in.

Sorry for jumping in shortly, very important here. That's the difference between the normal system we work on structured data. The assessments here are also better on structured data, but on the conformity checker, we can also here upload any format. That's very important to know. So if you have any tech talk in any format, we can make this assessment by sipping it in here.

Exactly. And looking a bit on the time, because I think that the idea is then pretty clear what's happening next. Right? AI then goes, sorts, I mean, you have seen it before, we uploaded 409 files to the system. They are analyzing them, crunching the numbers. Checking completeness against the main requirements you have to do. And then, we actually conduct reviews. And here you would see all the different areas where you have to be compliant. To… to receive the certification. And we can… Choose… Anyone we want. Let's have a look… the product realization planning. I'm very sure the one we uploaded is not compliant there.

And then, you have two options. You can either just start the agentic way, how I did it here, then the AI agent is really looking through all the files, trying to find the evidences. Really running here, and you can also see the reasoning of the agents step-by-step. And I also want to show here, for example. This is not a good one. For example, I think this is the one I ran before. So here you would see, this is a concrete assessment, which has to be done, and then the AI agent really goes through and tells us, I reviewed these documents, right? These were my thinkings, these documents are meeting the requirements, in these documents, you're having some gaps.

And then, you also get all your findings. And in the end, same, the agent is still running, like, it takes a lot of time, we're doing a lot of computation in the background to ensure that all of this is reliable and traceable. And when this is done, Then you will get… like a PDF, right? You then export yourself the audit report, it, like, renders this one, and I can show you how these things look in the end. Exactly like this. Oops.

Other one, Nico.

Yep, I'm sorry. Read your cycle one. That's the one I prepared before. Here. And then you also get complete conformity assessment reports, where you know exactly how did I do. on which categories? how… was, for example, this item evaluated, which evidences were reviewed for that, what were the findings, and then you also get corrective actions. And then these are the things you can feed in, again, to the data-driven system to update yourself, so you're in a constant loop of self-check, and this is also where the whole system has to go.

And maybe last words to this one? Thank you, Nico. What we couldn't show in this first demo is normally this is really strongly human in the loop, so we make this evidence, and we do the findings super fast, but then you can check every single analysis the system did, and can check yourself. Just because, for the sake of time, I would…

You could go more in deep and analyze and also ask more follow-up questions to these things, but the gist of being able to use AI agents to analyze your documents against requirements is really the gist of it. Yeah, and with this, maybe, back to you, Leon, to wrap it up, so we have also some time for questions. Where do I start?

I would say, to be honest, that's the very last slide. We have a short future outlook, but I would honestly prefer to bring in some questions now.

If there's anybody online that, wants to ask a question live. now would be the time to do so. In the meantime, I'm gonna say, you know, great job, guys. one of the things that I thought of a while ago, and you seem to have built it, is, sort of the ability to ask a notified body, what they think of your submission before you send them the submission, and… and this is… this is kind of like a digital notified body. This kind of, you know, doing the review that they're gonna do before they do the review, and so even though they cannot consult, you can. And

That gives you the ability to see where your submission is likely to get flagged, how it's gonna get, in trouble. Would you agree with that, or am I misunderstanding?

it's… it… let's say we… I would call it its max a pre-assessment, a pre-check before it, because obviously we need to have this clear separation between the worlds, and that's completely ensured, but having this conformity check or a simulated QM audit can really make the manufacturer be more secure in submitting, for example, and find also lags very early, so… Yes, you're correct. It's with limitations, obvious, what we are allowed to do and what is acceptable, but regarding this otherwise, yes, I agree with you.

Well, maybe also that you know where we see the ultimate dream state. would be people just work on the data, make changes, and you kind of always have, like, a traffic light, you know? As long as it's green, you know you're compliant, and then you make changes. And then you come in the next day, and it's like, you just updated something claiming a new patient population is there, but your clinical data is not supporting that. So, it's an orange, you are not ready. And then you have your gap analysis, you know what to do, and so you have, like, a real… this is where it needs to go, real time, or, like, shorter cycle of, like, checking if you're actually compliant and not developing for years.

Sending it to an authority, and then knowing where you stand. The cycles have to be way shorter, and reducing that is… It's why we're here, yeah.

Hey guys, I've watched a handful of presentations here over the last few years that had this much response from everybody, so I wanted to say, very nice job. How long have you been working on this?

Good question, sort of… 3 years now, I would say, so we started 2023, although openly speaking, we quitted our former jobs 2024 for that, and… Yeah, so you know what we do on the weekends.

Well, it's impressive for that short amount of time. Nice job.

Thank you.

So, we have a lot of consultants on the call. that work with clients that might be able to use your product. Do you have any sort of partnership program or ability for them to, work with you? And if so, how should they contact you? Right partner in the, In the chat, what… what is… what does that look like?

we partner a lot. Like, Leon, you can also maybe explain it, how the model is there. With the partnerships, but… but… or I do it as I'm talking already now. So, we also understand, right? It always comes down to the pyramid, what we said. So. people want to reach this nice digitalization layer, but sometimes they are internally not even ready, right? Like, you need to make the project management, or you already work together with partners who is your trusted peer. And especially for these projects, we really like working together with consultancies. We've done that successfully in the past, and then it's often…

we come in together as, like, partners, right? Like, people with expertise and the trust, and then we as a digitalization partner, and then you have successful projects together. And yeah, whoever is interested, I think the partner in the chat, I think, is really cool. So, we are very happy to reach out.

So, here, very simply, regarding partnerships, you can directly reach out to me in this case. For the manufacturer's side, you can also reach out directly to Benef also free assessment direction. Very important, it's like, we have these free ways of doing it. We have sometimes customers coming to us, they have super strong expertise in-house, they can use our system autonomously. it's, honestly speaking, the minority of the cases, and often we have companies who need a lot of support also on the first certification, and they're actually… we have now 25 partners around, and this works really, really well, this combination, so please reach out directly to me, and we can talk about it.

Yeah? And also as… yeah?

I would really like to thank you. I think this is fantastic, actually, and I… I mean, I personally, and I don't know… I can't speak for everybody else, but… there's a lot of thought behind this, and I see that you've got a lot of input from the industry already. My question is, if I may ask, is there any… I mean, is this a really fixed structure of… Of definitions, for example, like, if you wanted your intended purpose to appear in a different manner than what you presented today, is there a flexibility in the system? And I have a second question, the medical device knowledge units, I've been following that quite a bit. Are you the only ones using this in this type of manner, or do you have competitors? It's not sort of really a fair question, but I'm just kind of curious.

I can… I can answer both. So, on the first one, so MTKU, a lot of manufacturers themselves using the MDKAU standard and deem spec, especially it's quite German-driven, I would say. I would love to have this more international, more European-driven. And to the first one of your question, like, one… unique thing of our system is we have a complete modular data model, so we presented you all this, this is the data model, this is what you need, but if you go deeper into the topic, you can define it yourself. Why? Because we have, I don't know, in Europe, 35,000 manufacturers, and not… it doesn't fit for everyone. So you can use maybe 80% of the existing model, and 20% is specific to your device, and you can go

completely… Change this. Nico, just for maybe the sake of time.

Yeah, I'm so sorry, I just want to show. Everything you do, Every data model is… Just a configuration. Whenever you need something else, you just drop in new things, you say, I need a number here, all good. You just type it. So, when we started, after talking to the industry, they told us, when you try to make a one-size-fits-all, it will never work. So everything we do, all the templates, all the submissions, all the data model, it's completely flexible. We always… we give you the 80%, so you have something to start, and then the people build from there. But it's very, very important. Everything is flexible. I wanted to show that as well.

Technology-wise, that was actually one of the hardest things to build. That was maybe…

I have a big appreciation for this, because I worked, actually, on product data management, metadata, about 26 years ago, so I really understand what's behind it, and I'd like to applaud you. I think this is excellent, what you guys just showed us. I'm very impressed. Thank you.

And I'd like to thank you for joining us, and I'd like to thank everybody for sharing their time. Today, I don't know any other community online where you get to see new technologies like this, and meet, young innovators and, interact with them, and see what's coming, and find opportunities for yourself. So spread the word, join us next week for the InSilico Evidence, session, and, look for more exciting things to come, from CertHub. Thanks, guys.

Awesome.

Thank you very much.

Thanks, guys.

I'm…

Maria E. Donawa, MD: Well done.

Nice job.

Right?

Fantastic.

Thanks, Sean.

Thank you, thank you all.